基于颈椎CBCT三维影像分析构建南通地区女性青少年骨龄分期预测模型及验证  

Construction and validation of a bone age staging prediction model for female adolescents in Nantong area based on cervical CBCT three-dimensional imaging analysis

作　　者：赵桢祺 顾云 蔡颖[1] ZHAO Zhenqi;GU Yun;CAI Ying(Department of Stomatology,Nantong First People's Hospital,Nantong 226014,China)

机构地区：[1]南通市第一人民医院口腔科,江苏南通226014

出　　处：《口腔生物医学》2025年第2期94-100,共7页Oral Biomedicine

基　　金：南通儿童复杂口腔疾病诊疗资源优化对策建议(RB22-112)。

摘　　要：目的:探讨颈椎CBCT三维影像预测南通地区女性青少年骨龄分期的效果,开发并验证骨龄分期的预测模型。方法:回顾性分析2020年6月至2023年5月于南通市第一人民医院口腔正畸科就诊的南通地区女性青少年患者资料。随机抽样分为训练集和验证集,根据CVM分期法判断骨龄,并对患者进行分组,分为CVM1-CVM2组和CVM3-CVM5组。Logistic回归分析临床协变量,构建临床模型;进行组内相关系数(ICC)和线性相关性检验,建立LASSO回归验证筛选较优特征,支持向量机(SVM)构建影像组学标签和影像组学模型;将2种模型预测作为自变量纳入Logistic回归分析,构建联合模型;进行Boot-strap自抽样验证,ROC评估区分度,校准曲线评估准确度,决策曲线评估临床收益。结果:CVM3-CVM5的影响因素有年龄、锻炼情况、二维及三维形态参数、饮食习惯。筛选出影像组学标签10个特征(ICC及95%CI均>0.800),在CVM1-CVM2组和CVM3-CVM5组中有显著性差异(P<0.05)。联合模型在校准曲线分析中展现出优秀的拟合效果。联合模型(AUC:0.796;95%CI:0.712~0.819)敏感度为80.77%,特异度为84.03%,准确度为87.36%。此外,该模型的净收益显著高于其他影像学和临床特征。结论:颈椎CBCT三维影像组学联合临床变量构建的模型具有良好的临床可行性和临床决策价值。Objective:To investigate the study of cervical spine CBCT 3D imaging in predicting bone age stage of female adoles-cents in Nantong area,and to develop and validate a prediction model for bone age stage.Methods:The data of female adolescents in Nantong area who were treated in the Department of Orthodontics,Nantong First People's Hospital from June 2020 to May 2023 were retrospectively analyzed.The patients were randomly divided into training set and validation set.Bone age was judged according to the CVM staging method,and the patients were divided into CVM1-CVM2 group and CVM3-CVM5 group.Logistic regression was used to analyze clinical covariates and construct a clinical model.The intraclass correlation coefficient(ICC)and linear correlation test were used to establish LASSO regression to verify and select the best features,and support vector machine(SVM)was used to construct the radiomics signature and radiomics model.The two model predictions were included in Logistic regression analysis as independent varia-bles to construct a joint model.Bootstrap self-sampling verification,ROC evaluation of discrimination,calibration curve evaluation of accuracy,decision curve evaluation of clinical benefits.Results:The risk factors of CVM3-CVM5 were age,exercise status,two-di-mensional and three-dimensional morphological parameters,and dietary habits(P<0.05).10 features of radiomics signature(ICC and 95%CI>0.800)were selected,which showed significant differences between CVM1-CVM2 and CVM3-CVM5 groups(P<0.05).The joint model showed excellent fitting effect in the calibration curve analysis.The combined model(AUC:0.796,95%CI:0.712-0.819)had a sensitivity of 80.77%,a specificity of 84.03%,and an accuracy of 87.36%.In addition,the net benefit of the model was signifi-cantly higher than those for other imaging and clinical features.Conclusions:The model of cervical spine CBCT three-dimensional ra-diomics combined with clinical variables has good clinical feasibility and clinical decision-making value.

关 键 词：锥形束计算机断层扫描 骨龄分期 颈椎 模型预测 

分 类 号：R783.5[医药卫生—口腔医学]